The growing number of complicated fluid devices and rising challenges of complex turbulent flows across numerous disciplines necessitate the development of robust multiscale variable fidelity computational methodologies. A wide range of hybrid approaches have been attempted in order to smooth the sharp interface between the different existing models, namely, Reynolds-Averaged Navier-Stokes (RANS), Large Eddy Simulation (LES), and Direct Numerical Simulation (DNS). A conceptual problem with most current hybrid approaches is that they are based on the common perception that turbulence modeling and numerical methods are two separate fields of research and that, once a turbulence model is developed, any suitable computational approach can be used for its numerical implementation.  In this talk a novel framework for continuously variable fidelity space/time/model-form adaptive turbulence simulation that tightly integrates numerics and modeling to ensure better capturing of the flow physics on a near optimal adaptive mesh is presented. The integration is achieved by combining spatially and temporally varying wavelet thresholding with hierarchical wavelet-based turbulence modeling. The resulting approach provides automatic smooth transition from directly resolving all flow physics to capturing only the energetic/coherent structures, leading to a dynamically adaptive variable fidelity approach. The self-regulating continuous switch between different fidelity regimes is accomplished through spatiotemporal variation of the wavelet threshold and two-way feedback mechanism between the modeled dissipation and the local grid resolution. This defines a new concept of model-refinement. The new methodology of physics-based spatially varying thresholding fully exploits the spatial/temporal intermittency of turbulence resulting in considerably better Reynolds number scaling, which is independent of the fidelity of the simulations. The ability of the proposed methodology to capture the flow-physics at the desired level of fidelity is demonstrated for the benchmark problem where the fidelity of turbulence simulation, measured by the ratio of the SGS and total dissipations, automatically adjusts to time-varying user prescribed levels. Finally, the implementation of the proposed model-refinement concept within classical LES methodology and possible feedback mechanism to incorporate a filter-width/model adaptation, preferably coupled with adaptation of the numerical resolution, are discussed.